GIS and Remote Sensing Assessment of Landslide Susceptibility along the Cameroon Volcanic Line: West Region, Cameroon

The physical and tectonic setting exposes the western part of Cameroon to natural and anthropogenic hazards. Small scale landslides with devastating effects are recurrent along the Cameroon Volcanic Line. Limited studies have addressed the susceptibility to sliding in the area. This study therefore aimed at producing a landslide susceptibility map of the West Region to aid local and national authorities in land use planning and policy to minimise loss. Eleven conditioning and triggering factors were selected to investigate landslide susceptibility in the study area. These factors include; slope angle, lithology, soil, slope aspect, elevation, rainfall, geological faults, land use, normalised difference vegetation index, roads and river networks. These factors were assigned weights using the analytical hierarchy process. The weighted linear combination technique was used to derive landslide susceptibility indices and the susceptibility map. The map was reclassied into ve classes; very low, low, moderate, high and very susceptibility class. About 16% (2180 km 2 ) of the study area lies within the high to very high class while 47% (6512 km 2 ) is found within the moderate class. Steep slopes, weathered volcanic rocks and thick soil cover at high elevations control the distribution of landslides while high intensity rainfall is the main triggering factor. Residential houses and road infrastructures along steep slopes are the most vulnerable to sliding. Site specic assessment needs to be conducted in order to implement effective mitigation measures.


Introduction
The physical and tectonic setting of Cameroon coupled with changing climatic conditions exposes the western part of the country to both natural and anthropogenic hazards (Ayonghe et al. 1999(Ayonghe et al. , 2002(Ayonghe et al. , 2004 produced a landslide susceptibility map for Limbe (a section of the tectonically active Cameroon Volcanic Line), the recent landslide in Gouache neighbourhood in Bafoussam (Fig. 1c), an area with no historical landslide record sheds light on the necessity to map the entire region. This study is therefore aimed at producing a landslide hazard zonation map that will be used by local and national authorities for land use planning and policy to minimize loss.The objectives of this work are; to identify areas more likely to be affected by landslides in the future using satellite images and the analytical hierarchy process, to understand the factors leading to slope instability in the region from the multicriteria decision analysis, to identify settlements and land uses in the high-risk zone for possible relocation or preventive measures, to increase awareness and add to the already growing data on hazards in Cameroon.

Study Area
The study area is situated in the geopolitical West Region of Cameroon within longitude 10 o 30' 00" and latitude 5 o 30' 00" (Fig. 1c). It is the smallest of the ten regions of Cameroon with a surface area of approximately 13892 km 2 , a total population of 1,921,590 as of 2015 and a high population density of 140/km 2 (Tesi, 2018). Bafoussam, the political capital of the West Region is located about 336 km from the national capital, Yaounde. The area has a moderate Equatorial climate resulting from high elevation and high humidity. Temperatures vary between 15 o C to 28 o C. This region experiences high rainfall averaging 1000-2000 mm/year (Tesi, 2018).
The topography of the West Region is generally mountainous with elevations ranging from 224 to 2744 metres above sea level. As a result of the mountainous terrain, fast-owing rivers are ubiquitous. Several crater lakes have developed from collapsed volcanoes. The area has a variable soil type comprising of ferralitic soils and alluvial soils derived from the weathering of plutonic and volcanic rocks.
The original forest vegetation has been cleared for agriculture giving rise to grassland vegetation.
Patches of Woodland Savannah of the Sahel type are found distributed within the area. Plantation farming is practised on a small scale, with coffee, cocoa, tea and tobacco as the main cash crops.
Livestock farming includes; cattle, sheep and goat rearing. Poultry and piggery farming is also increasingly practised in recent years. The region is well known for artistic and craftsmanship which involves the production of high-quality ceramics from clay, woodworks, brass and bronze casting and cotton textiles (Tesi, 2018

Data Sources for Landslide Conditioning Factors
Various factors have been selected to investigate landslide susceptibility in the study area. Table 1 details the type and source of data used in the assessment of landslides susceptibility. Thematic maps were generated from these data in a GIS environment. Given the paucity of literature in the study area, the Landsat 8 operational land imager (OLI) images were downloaded from the United States Geological Survey website (Table 1). These images were layer stacked in ERDAS Imagine 2018 employing contrast enhancement and feathering techniques (Kumar 2005).

Multicriteria Decision Analysis
Multicriteria decision analysis (MCDA) is a GIS-based method for decision making through the integration of geographic data and subjective judgements (Malczewski 1999).

Analytical Hierarchy Process
An analytic hierarchy process (AHP) is a form of MCDA quantitative method for decision making using factor weights through pairwise comparison (Saaty 1987). This method measures both tangible and intangible variables through relative weights given to each variable based on the preference of the researcher. It has been widely applied in MCDA, planning, natural and man-made resource allocation, and con ict resolution (Saaty 1986 The AHP method has three distinct facets; decomposition, comparative judgment and synthesis of priorities. A complex problem is broken down into a hierarchy of variables or factors using a pairwise comparison matrix, factors are assigned weights on a nine-point scale see Table 2 (Malczewski 1999;Ahmed 2015). The pairwise comparison is based on two intrinsic questions to determine criterion or factor more important than the others and the extent based on a ratio scale of 1/9 to 9 ( Table 2). The AHP calculation was undertaken using Microsoft excel.
To validate the results of the pairwise comparison metrics and factor weights, the consistency index (CI) and the consistency ratio (CR) was determined (Eastman 2012). The consistency index is given by Where CI = consistency index λmax = normalized highest Eigenvalue of the pairwise matrix n = number of factors (11 factors in this study) The consistency ratio shows how random the matrix ratings were selected as given by Saaty (1980).
Random index (RI) has been proposed by Saaty (1987) and presented in Table 3 A consistency ratio of 0 implies perfect ratings of factors, CR of > 0.1 implies inconsistency of the ratings. Saaty (1980) suggested a re-evaluation of factor ratings for CR > 0.1.
The result of a pairwise comparison matrix gives rise to factor weight which is then aggregated to generate a landslide susceptibility map ( Landslide inventory map can be used as a means for assigning weights to landslide triggering factors (Kumar et al. 2018). These maps can be generated from aerial photographs, eld surveys, satellite images and existing landslides. Fourteen landslides were determined in the study area from the review of literature (Table 4) and the classi cation of satellite images (Fig. 1c).

Land use and Normalized Difference Vegetation Index
Land use map was generated from the Landsat 8 OLI satellite image through supervised classi cation using maximum likelihood (Lu and Weng 2007). False-colour composite images and Google Earth were used to obtain training data through the polygon method. Five landcover classes were identi ed; water body, agricultural land, built-up area, vegetation and bare soil (Fig. 2a).
Due to the in uence of vegetation coverage on slope stability, normalized difference vegetation index (NDVI) was carried out to characterize vegetation extent in the study area Eq. 4 Where NDVI = normalized difference vegetation index NDVI analysis results in an output of values ranging from − 1 to 1 where the negative values represent clouds, water and snow (Zaitunah et al. 2018). NDVI values of 0-0.1 represent barren land, rocks and soils while values of 0.6-1 represent dense vegetation (Fig. 2b).

Elevation
A 30m resolution shuttle radar topography mission (SRTM) digital elevation model (DEM) was downloaded from the USGS website. The average elevation of the study area is 155m, the lowest point is 224m and the highest point is 2744m (Fig. 2c). Generally, areas with higher elevations are more susceptible to landslides. The elevation generated was reclassi ed into ve classes to determine the level of contribution of each category to landslides.

Slope
The digital elevation model (DEM) was used to generate a slope map, the slope in the study area ranges

Aspect
Aspect refers to the orientation of a slope from 0 o to 360 o . Sunlight exposure, drying winds, rainfall and discontinuities are factors associated with slope aspect which in uences the degree of susceptibility to landslides (Dai et al. 2001). Nine slope directions were generated and reclassi ed according to their contribution to landslide susceptibility (Fig. 2e).

Geology
A scanned geologic map of Cameroon was georeferenced, the geology of the study area was digitised into polygons that were converted to raster format. Four lithologies were identi ed; pre-syn tectonic granitoids, syn-post tectonic granitoids, orthogneiss, and volcanic rocks (Fig. 2f). The area has highly weathered volcanic rocks which have been identi ed in some studies as landslide-prone lithologies (Che et al. 2011).

Soils
The stability of slopes depends on the soils they contain (Sartohadi et al. 2018; Schiliro et al. 2019). The soil map was digitized from the African groundwater Atlas map. The soil atlas was converted from shape le to a 30 m raster le in Arcmap. Five soil types of varying permeability and susceptibility to landslides were derived; andosols, loxisols, luvisols, stagnosols, and vertisols (Fig. 2g). luvisols are the dominant soil type in the study area. Soils capable of holding water have a higher level of susceptibility (Nandi and Shakoor 2009).

Rainfall
The average monthly rainfall data of the study area from the year 2000 to 2020 were downloaded from the NASA Earth Data website, this data was interpolated to generate the rainfall map (Fig. 2h). The average monthly rainfall ranges from 97 to 171mm/month. The rainfall data were reclassi ed into ve classes representing susceptibility levels. repository as shape les. The shape les were converted to raster data with a resolution of 30m. The Euclidean distance function in ArcMap was used to derive the distance to roads and rivers. Five classes were generated for both distances to road and distance to rivers ( Fig. 2i and j).

Distance to Fault
From the georeferenced geologic map of Cameroon, faults were digitised into line features (Fig. 2k). The multiple ring buffer function was used to generate a distance to faults. Five classes were derived with intervals of 5km.

Aggregation of Factor weights
In this study, the weighted linear combination (WLC) method was used (Appendix). This method is customized in many GIS platforms and it is exible in combining thematic maps of conditioning factors to generate landslide susceptibility maps (Feizizadeh and Blaschke 2013). It requires the standardization of classes within each factor to a common numeric scale. The factor classes are multiplied by the weights obtained from the comparison matrix and their results summed to obtain the landslide susceptibility index (Eq. 3).
Where LSI = Landslide Susceptibility Index Wj = Weight value of causative factor j Zij = Weight value of class i of causative factor j The landslide susceptibility indices generated was reclassi ed to derive the landslide susceptibility map using the Jenk classi cation method. The map was reclassi ed into ve classes; very high, high, moderate, low and very low susceptibilities (Fig. 3).

Landslide susceptibility Map
Landslide susceptibility map was generated through the weighted linear combination method using 11 landslide conditioning factors (Fig. 3, Table 5). The landslide susceptibility map was classi ed into ve classes using the natural break (Jenk) method; very low, low, moderate, high and very high susceptibility. The Jenk method is a data clustering technique used to determine the best grouping of values into different classes by minimising the deviation of each class from the class mean while maximising the class deviation from the means of other groups (Raja et al. 2017).
To create a landslide susceptibility map using the AHP technique, a pairwise comparison matrix is constructed. The matrix is used in assigning factor ratings and for calculating factor weights (Tables 5  and 6). The consistency ratio determines the degree of consistency in assigning the factor weights. In this study, the consistency ratio (CR) is < 0.1.
The area and percentage coverage of ve landslide classes is presented in Table 7. From Table 7, the medium landslide category has the highest area coverage of 6512 km 2 (47%) followed by the low category 3149 km 2 (23%). The very low landslide category occupies 2051 km 2 (15%). The lowest area coverage is occupied by the high landslide category 230 km 2 (2%). The very high landslide category covers 1950 km 2 (14%) of the study area. Therefore, about 16% (2180 km 2 ) of the study area falls within the high and very high landslide category (Table 7).

Landslide Model validation
To determine how successful the model is in predicting landslide susceptible zones, the landslide inventory was superimposed on the landslide susceptibility map see Table 7. From Table 7 43% of landslide inventory falls within the high to very high susceptibility class. About 36% of landslide falls within the medium class while 21% falls within the low susceptibility class. The result shows that no landslide has been recorded in the very low landslide susceptibility class. The result obtained is representative of the study area when compared to the landslide inventory.

Uncertainties in landslide model assessment
In the MCDA using the AHP technique, errors may result from assigning incorrect factor weights. For example, Ahmed (2015) noted that high susceptibility class was found at low elevation which is not normally associated with landslides. This resulted from the assignment of weight to some factors which occurred both in high as well as low elevations. Therefore, results obtained using MCDA may have inherent errors. As a result of the subjectivity in assigning factor weights, an incorrect weight assignment affects the accuracy of the susceptibility map generated (Kritikos and Davies 2011). Another source of error is the incorrect pairwise comparison of factors. Ahmed (2015) suggested that different combination of factors be taken into consideration in order to derive the most appropriate factor weights. To ensure appropriate rating and weighting of factors, Saaty (1980) and Eastman (2012) developed the consistency index and consistency ratio for determining the randomness in assigning factor weights which in turn affects the quality of the susceptibility map generated. The consistency ratio for both factors and subfactors is less than 0.1 (Tables 5 and 6) which is the cut-off set by Saaty (1980) for revising factor weighting. Therefore, the susceptibility map created in this study is of good quality.
The quality of the data set used may also introduce errors in the map generated. The spatial resolution of satellite images and SRTM DEM constraints its effectiveness in studying landslides which also affects (2002) were successful in identifying just 25% of landslides in Nepal's highlands using Landsat 7 ETM + images with a 30 m spatial resolution pansharpened to 15 m. Nichol and Wong (2005) showed that at a 1m spatial resolution, satellite images may not be suitable for identifying small scale landslides (< 10 m wide). Therefore, a 30m spatial resolution satellite image and DEM is not suitable for this study given that most of the landslides along the CVL are of small scale (Zogning et al. 2007;Thierry et al. 2008;Zangmo et al. 2009;Wotchoko et al. 2016). However, this is the data set freely available for the study.
When georeferencing and digitising raster data, errors are usually introduced (Pearson 2006). To reduce the error margins, ground control points were carefully selected and a rst order polynomial was used to georeference the geologic, faults and soil maps. Topological rules (no overlapping, no dangle lines and no pseudo lines) were used to ensure accurate digitising of faults, geology and soil maps (Pearson 2006). The rainfall data obtained from the NASA Earth Data website was interpolated using the kriging technique (Ly et al. 2013). An averaging algorithm was used to ll in the missing time gap data (Acker and Leptoukh 2007). Data acquired from Open Street Map (OSM) may be inaccurate resulting from the contribution of data from volunteers with little to no geospatial knowledge. However, the availability of high-resolution aerial and satellite images enables e cient data collection and validation of digitised features (OSM 2020).

Risk Assessment
The ultimate aim for undertaking a landslide susceptibility assessment is to determine land-use types that might be affected in an event of a landslide in order to implement mitigation measures (Varness 1984; Awawdeh et al. 2018). A qualitative approach was used to determine the risk associated with landslide categories in agricultural land, urban area and road infrastructures. Some roads in the study area are found within high-risk zone, for example, about 42 km of the Bafoussam-Foumban highway lies within the very high susceptibility class while 22 km passes along the high susceptibility class (Fig. 4).
Similarly, the national road (N4) running from Douala-Bafoussam cuts across the moderate susceptibility class for over 58 km and the high category for about 6 km (Fig. 4). Furthermore, the Mbouda-Bamenda highway passes through the moderate susceptibility class for over 48 km.
Results show that about 10% of the urban area comprising of parts of Foumban, Foumbot, Bafoussam, Santchou, and Mbouda falls within the very high class (Fig. 4). About 7% of the urban area including parts of Bafoussam, Bangante, Santchou lies within the high category. Over 30% of the urban area is located in the moderate susceptibility class. Residential houses built along unstable slopes have been identi ed as the most vulnerable to sliding. Another important land use which might be affected by landslides is agricultural land. Over 50% of agricultural land is found within the moderate susceptibility category. This has been identi ed in some studies as a landslide inducing factor as it involves the cutting inevitable. This is exacerbated by changing climatic conditions leading to increase in rainfall regime. This study therefore aimed at undertaking a landslide susceptibility assessment to aid local and national authorities in land use and policy planning to minimise the destructive effects of landslides. Eleven landslide conditioning factors were selected to investigate slope stability in the study area. Steep slopes, high elevations, weathered volcanic rocks and thick soil cover along steep slopes are identi ed as the major landslide causative factors. High intensity rainfall (110 mm/day) for over a period of 2-4 days is the main landslide triggering factor in the area. The landslide susceptibility map generated from the multicriteria decision analysis was subdivided into ve classes; very low, low, moderate, high and very high susceptibility class. Roads and residential houses built along steep slopes are the most vulnerable infrastructures to slides. Agricultural land is less vulnerable to sliding as over 50% is found within the moderate susceptibility class. To curb the effect of landslides, some mitigation mechanism such as; tree planting, slope terracing, draining pipes, construction of retaining walls is proposed. In order to implement appropriate mitigation measures, it is recommended that site speci c assessments be conducted to identify triggering factors in high to very susceptibility zones.

Declarations
Funding No funding was received to assist with the preparation of this manuscript.

Con ict of interest
The author declares no con ict of interest Availability of data Code availability 'Not applicable' Author's contribution This work is an output from the author's MSc dissertation. It was conceived and produced by the author.     Roads and urban areas in high to very high landslide susceptibility zones

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.